That’s actually a part of the propolys-tts.agi file. For some TTS engines it generates the .sln file directly from the engine. For others (including pico), it does two steps: first generating a .wav file from the TTS engine, then converting it to the required output type using sox.
That’s why the filename of the .wav file has -temp- in it.
As I noted in the original post, the AGI does detect this intermediate .wav file if I manually generate it on the command line, and then processes it to .sln and plays the audio.